1 /*-
2 * modified for Lites 1.1
3 *
4 * Aug 1995, Godmar Back (gback@cs.utah.edu)
5 * University of Utah, Department of Computer Science
6 */
7 /*-
8 * Copyright (c) 1989, 1993
9 * The Regents of the University of California. All rights reserved.
10 * (c) UNIX System Laboratories, Inc.
11 * All or some portions of this file are derived from material licensed
12 * to the University of California by American Telephone and Telegraph
13 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
14 * the permission of UNIX System Laboratories, Inc.
15 *
16 * Redistribution and use in source and binary forms, with or without
17 * modification, are permitted provided that the following conditions
18 * are met:
19 * 1. Redistributions of source code must retain the above copyright
20 * notice, this list of conditions and the following disclaimer.
21 * 2. Redistributions in binary form must reproduce the above copyright
22 * notice, this list of conditions and the following disclaimer in the
23 * documentation and/or other materials provided with the distribution.
24 * 4. Neither the name of the University nor the names of its contributors
25 * may be used to endorse or promote products derived from this software
26 * without specific prior written permission.
27 *
28 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * SUCH DAMAGE.
39 *
40 * @(#)ufs_lookup.c 8.6 (Berkeley) 4/1/94
41 * $FreeBSD$
42 */
43
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/namei.h>
47 #include <sys/bio.h>
48 #include <sys/buf.h>
49 #include <sys/endian.h>
50 #include <sys/mount.h>
51 #include <sys/vnode.h>
52 #include <sys/malloc.h>
53 #include <sys/dirent.h>
54 #include <sys/sysctl.h>
55
56 #include <ufs/ufs/dir.h>
57
58 #include <fs/ext2fs/inode.h>
59 #include <fs/ext2fs/ext2_mount.h>
60 #include <fs/ext2fs/ext2fs.h>
61 #include <fs/ext2fs/ext2_dinode.h>
62 #include <fs/ext2fs/ext2_dir.h>
63 #include <fs/ext2fs/ext2_extern.h>
64
65 #ifdef INVARIANTS
66 static int dirchk = 1;
67 #else
68 static int dirchk = 0;
69 #endif
70
71 static SYSCTL_NODE(_vfs, OID_AUTO, e2fs, CTLFLAG_RD, 0, "EXT2FS filesystem");
72 SYSCTL_INT(_vfs_e2fs, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0, "");
73
74 /*
75 DIRBLKSIZE in ffs is DEV_BSIZE (in most cases 512)
76 while it is the native blocksize in ext2fs - thus, a #define
77 is no longer appropriate
78 */
79 #undef DIRBLKSIZ
80
81 static u_char ext2_ft_to_dt[] = {
82 DT_UNKNOWN, /* EXT2_FT_UNKNOWN */
83 DT_REG, /* EXT2_FT_REG_FILE */
84 DT_DIR, /* EXT2_FT_DIR */
85 DT_CHR, /* EXT2_FT_CHRDEV */
86 DT_BLK, /* EXT2_FT_BLKDEV */
87 DT_FIFO, /* EXT2_FT_FIFO */
88 DT_SOCK, /* EXT2_FT_SOCK */
89 DT_LNK, /* EXT2_FT_SYMLINK */
90 };
91 #define FTTODT(ft) \
92 ((ft) < nitems(ext2_ft_to_dt) ? ext2_ft_to_dt[(ft)] : DT_UNKNOWN)
93
94 static u_char dt_to_ext2_ft[] = {
95 EXT2_FT_UNKNOWN, /* DT_UNKNOWN */
96 EXT2_FT_FIFO, /* DT_FIFO */
97 EXT2_FT_CHRDEV, /* DT_CHR */
98 EXT2_FT_UNKNOWN, /* unused */
99 EXT2_FT_DIR, /* DT_DIR */
100 EXT2_FT_UNKNOWN, /* unused */
101 EXT2_FT_BLKDEV, /* DT_BLK */
102 EXT2_FT_UNKNOWN, /* unused */
103 EXT2_FT_REG_FILE, /* DT_REG */
104 EXT2_FT_UNKNOWN, /* unused */
105 EXT2_FT_SYMLINK, /* DT_LNK */
106 EXT2_FT_UNKNOWN, /* unused */
107 EXT2_FT_SOCK, /* DT_SOCK */
108 EXT2_FT_UNKNOWN, /* unused */
109 EXT2_FT_UNKNOWN, /* DT_WHT */
110 };
111 #define DTTOFT(dt) \
112 ((dt) < nitems(dt_to_ext2_ft) ? dt_to_ext2_ft[(dt)] : EXT2_FT_UNKNOWN)
113
114 static int ext2_dirbadentry(struct vnode *dp, struct ext2fs_direct_2 *de,
115 int entryoffsetinblock);
116 static int ext2_is_dot_entry(struct componentname *cnp);
117 static int ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp,
118 struct componentname *cnp, ino_t *dd_ino);
119
120 static int
121 ext2_is_dot_entry(struct componentname *cnp)
122 {
123 if (cnp->cn_namelen <= 2 && cnp->cn_nameptr[0] == '.' &&
124 (cnp->cn_nameptr[1] == '.' || cnp->cn_nameptr[1] == '\0'))
125 return (1);
126 return (0);
127 }
128
129 /*
130 * Vnode op for reading directories.
131 */
132 int
133 ext2_readdir(struct vop_readdir_args *ap)
134 {
135 struct vnode *vp = ap->a_vp;
136 struct uio *uio = ap->a_uio;
137 struct buf *bp;
138 struct inode *ip;
139 struct ext2fs_direct_2 *dp, *edp;
140 u_long *cookies;
141 struct dirent dstdp;
142 off_t offset, startoffset;
143 size_t readcnt, skipcnt;
144 ssize_t startresid;
145 int ncookies;
146 int DIRBLKSIZ = VTOI(ap->a_vp)->i_e2fs->e2fs_bsize;
147 int error;
148
149 if (uio->uio_offset < 0)
150 return (EINVAL);
151 ip = VTOI(vp);
152 if (ap->a_ncookies != NULL) {
153 if (uio->uio_resid < 0)
154 ncookies = 0;
155 else
156 ncookies = uio->uio_resid;
157 if (uio->uio_offset >= ip->i_size)
158 ncookies = 0;
159 else if (ip->i_size - uio->uio_offset < ncookies)
160 ncookies = ip->i_size - uio->uio_offset;
161 ncookies = ncookies / (offsetof(struct ext2fs_direct_2,
162 e2d_namlen) + 4) + 1;
163 cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
164 *ap->a_ncookies = ncookies;
165 *ap->a_cookies = cookies;
166 } else {
167 ncookies = 0;
168 cookies = NULL;
169 }
170 offset = startoffset = uio->uio_offset;
171 startresid = uio->uio_resid;
172 error = 0;
173 while (error == 0 && uio->uio_resid > 0 &&
174 uio->uio_offset < ip->i_size) {
175 error = ext2_blkatoff(vp, uio->uio_offset, NULL, &bp);
176 if (error)
177 break;
178 if (bp->b_offset + bp->b_bcount > ip->i_size)
179 readcnt = ip->i_size - bp->b_offset;
180 else
181 readcnt = bp->b_bcount;
182 skipcnt = (size_t)(uio->uio_offset - bp->b_offset) &
183 ~(size_t)(DIRBLKSIZ - 1);
184 offset = bp->b_offset + skipcnt;
185 dp = (struct ext2fs_direct_2 *)&bp->b_data[skipcnt];
186 edp = (struct ext2fs_direct_2 *)&bp->b_data[readcnt];
187 while (error == 0 && uio->uio_resid > 0 && dp < edp) {
188 if (dp->e2d_reclen <= offsetof(struct ext2fs_direct_2,
189 e2d_namlen) || (caddr_t)dp + dp->e2d_reclen >
190 (caddr_t)edp) {
191 error = EIO;
192 break;
193 }
194 /*-
195 * "New" ext2fs directory entries differ in 3 ways
196 * from ufs on-disk ones:
197 * - the name is not necessarily NUL-terminated.
198 * - the file type field always exists and always
199 * follows the name length field.
200 * - the file type is encoded in a different way.
201 *
202 * "Old" ext2fs directory entries need no special
203 * conversions, since they are binary compatible
204 * with "new" entries having a file type of 0 (i.e.,
205 * EXT2_FT_UNKNOWN). Splitting the old name length
206 * field didn't make a mess like it did in ufs,
207 * because ext2fs uses a machine-independent disk
208 * layout.
209 */
210 dstdp.d_namlen = dp->e2d_namlen;
211 dstdp.d_type = FTTODT(dp->e2d_type);
212 if (offsetof(struct ext2fs_direct_2, e2d_namlen) +
213 dstdp.d_namlen > dp->e2d_reclen) {
214 error = EIO;
215 break;
216 }
217 if (offset < startoffset || dp->e2d_ino == 0)
218 goto nextentry;
219 dstdp.d_fileno = dp->e2d_ino;
220 dstdp.d_reclen = GENERIC_DIRSIZ(&dstdp);
221 bcopy(dp->e2d_name, dstdp.d_name, dstdp.d_namlen);
222 dstdp.d_name[dstdp.d_namlen] = '\0';
223 if (dstdp.d_reclen > uio->uio_resid) {
224 if (uio->uio_resid == startresid)
225 error = EINVAL;
226 else
227 error = EJUSTRETURN;
228 break;
229 }
230 /* Advance dp. */
231 error = uiomove((caddr_t)&dstdp, dstdp.d_reclen, uio);
232 if (error)
233 break;
234 if (cookies != NULL) {
235 KASSERT(ncookies > 0,
236 ("ext2_readdir: cookies buffer too small"));
237 *cookies = offset + dp->e2d_reclen;
238 cookies++;
239 ncookies--;
240 }
241 nextentry:
242 offset += dp->e2d_reclen;
243 dp = (struct ext2fs_direct_2 *)((caddr_t)dp +
244 dp->e2d_reclen);
245 }
246 bqrelse(bp);
247 uio->uio_offset = offset;
248 }
249 /* We need to correct uio_offset. */
250 uio->uio_offset = offset;
251 if (error == EJUSTRETURN)
252 error = 0;
253 if (ap->a_ncookies != NULL) {
254 if (error == 0) {
255 ap->a_ncookies -= ncookies;
256 } else {
257 free(*ap->a_cookies, M_TEMP);
258 *ap->a_ncookies = 0;
259 *ap->a_cookies = NULL;
260 }
261 }
262 if (error == 0 && ap->a_eofflag)
263 *ap->a_eofflag = ip->i_size <= uio->uio_offset;
264 return (error);
265 }
266
267 /*
268 * Convert a component of a pathname into a pointer to a locked inode.
269 * This is a very central and rather complicated routine.
270 * If the file system is not maintained in a strict tree hierarchy,
271 * this can result in a deadlock situation (see comments in code below).
272 *
273 * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
274 * on whether the name is to be looked up, created, renamed, or deleted.
275 * When CREATE, RENAME, or DELETE is specified, information usable in
276 * creating, renaming, or deleting a directory entry may be calculated.
277 * If flag has LOCKPARENT or'ed into it and the target of the pathname
278 * exists, lookup returns both the target and its parent directory locked.
279 * When creating or renaming and LOCKPARENT is specified, the target may
280 * not be ".". When deleting and LOCKPARENT is specified, the target may
281 * be "."., but the caller must check to ensure it does an vrele and vput
282 * instead of two vputs.
283 *
284 * Overall outline of ext2_lookup:
285 *
286 * search for name in directory, to found or notfound
287 * notfound:
288 * if creating, return locked directory, leaving info on available slots
289 * else return error
290 * found:
291 * if at end of path and deleting, return information to allow delete
292 * if at end of path and rewriting (RENAME and LOCKPARENT), lock target
293 * inode and return info to allow rewrite
294 * if not at end, add name to cache; if at end and neither creating
295 * nor deleting, add name to cache
296 */
297 int
298 ext2_lookup(struct vop_cachedlookup_args *ap)
299 {
300
301 return (ext2_lookup_ino(ap->a_dvp, ap->a_vpp, ap->a_cnp, NULL));
302 }
303
304 static int
305 ext2_lookup_ino(struct vnode *vdp, struct vnode **vpp, struct componentname *cnp,
306 ino_t *dd_ino)
307 {
308 struct inode *dp; /* inode for directory being searched */
309 struct buf *bp; /* a buffer of directory entries */
310 struct ext2fs_direct_2 *ep; /* the current directory entry */
311 int entryoffsetinblock; /* offset of ep in bp's buffer */
312 struct ext2fs_searchslot ss;
313 doff_t i_diroff; /* cached i_diroff value */
314 doff_t i_offset; /* cached i_offset value */
315 int numdirpasses; /* strategy for directory search */
316 doff_t endsearch; /* offset to end directory search */
317 doff_t prevoff; /* prev entry dp->i_offset */
318 struct vnode *pdp; /* saved dp during symlink work */
319 struct vnode *tdp; /* returned by VFS_VGET */
320 doff_t enduseful; /* pointer past last used dir slot */
321 u_long bmask; /* block offset mask */
322 int error;
323 struct ucred *cred = cnp->cn_cred;
324 int flags = cnp->cn_flags;
325 int nameiop = cnp->cn_nameiop;
326 ino_t ino, ino1;
327 int ltype;
328 int entry_found = 0;
329
330 int DIRBLKSIZ = VTOI(vdp)->i_e2fs->e2fs_bsize;
331
332 if (vpp != NULL)
333 *vpp = NULL;
334
335 dp = VTOI(vdp);
336 bmask = VFSTOEXT2(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
337 restart:
338 bp = NULL;
339 ss.slotoffset = -1;
340
341 /*
342 * We now have a segment name to search for, and a directory to search.
343 *
344 * Suppress search for slots unless creating
345 * file and at end of pathname, in which case
346 * we watch for a place to put the new file in
347 * case it doesn't already exist.
348 */
349 i_diroff = dp->i_diroff;
350 ss.slotstatus = FOUND;
351 ss.slotfreespace = ss.slotsize = ss.slotneeded = 0;
352 if ((nameiop == CREATE || nameiop == RENAME) &&
353 (flags & ISLASTCN)) {
354 ss.slotstatus = NONE;
355 ss.slotneeded = EXT2_DIR_REC_LEN(cnp->cn_namelen);
356 /*
357 * was ss.slotneeded = (sizeof(struct direct) - MAXNAMLEN +
358 * cnp->cn_namelen + 3) &~ 3;
359 */
360 }
361 /*
362 * Try to lookup dir entry using htree directory index.
363 *
364 * If we got an error or we want to find '.' or '..' entry,
365 * we will fall back to linear search.
366 */
367 if (!ext2_is_dot_entry(cnp) && ext2_htree_has_idx(dp)) {
368 numdirpasses = 1;
369 entryoffsetinblock = 0;
370 switch (ext2_htree_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
371 &bp, &entryoffsetinblock, &i_offset, &prevoff,
372 &enduseful, &ss)) {
373 case 0:
374 ep = (struct ext2fs_direct_2 *)((char *)bp->b_data +
375 (i_offset & bmask));
376 goto foundentry;
377 case ENOENT:
378 i_offset = roundup2(dp->i_size, DIRBLKSIZ);
379 goto notfound;
380 default:
381 /*
382 * Something failed; just fallback to do a linear
383 * search.
384 */
385 break;
386 }
387 }
388
389 /*
390 * If there is cached information on a previous search of
391 * this directory, pick up where we last left off.
392 * We cache only lookups as these are the most common
393 * and have the greatest payoff. Caching CREATE has little
394 * benefit as it usually must search the entire directory
395 * to determine that the entry does not exist. Caching the
396 * location of the last DELETE or RENAME has not reduced
397 * profiling time and hence has been removed in the interest
398 * of simplicity.
399 */
400 if (nameiop != LOOKUP || i_diroff == 0 ||
401 i_diroff > dp->i_size) {
402 entryoffsetinblock = 0;
403 i_offset = 0;
404 numdirpasses = 1;
405 } else {
406 i_offset = i_diroff;
407 if ((entryoffsetinblock = i_offset & bmask) &&
408 (error = ext2_blkatoff(vdp, (off_t)i_offset, NULL,
409 &bp)))
410 return (error);
411 numdirpasses = 2;
412 nchstats.ncs_2passes++;
413 }
414 prevoff = i_offset;
415 endsearch = roundup2(dp->i_size, DIRBLKSIZ);
416 enduseful = 0;
417
418 searchloop:
419 while (i_offset < endsearch) {
420 /*
421 * If necessary, get the next directory block.
422 */
423 if (bp != NULL)
424 brelse(bp);
425 error = ext2_blkatoff(vdp, (off_t)i_offset, NULL, &bp);
426 if (error != 0)
427 return (error);
428 entryoffsetinblock = 0;
429 /*
430 * If still looking for a slot, and at a DIRBLKSIZE
431 * boundary, have to start looking for free space again.
432 */
433 if (ss.slotstatus == NONE &&
434 (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
435 ss.slotoffset = -1;
436 ss.slotfreespace = 0;
437 }
438 error = ext2_search_dirblock(dp, bp->b_data, &entry_found,
439 cnp->cn_nameptr, cnp->cn_namelen,
440 &entryoffsetinblock, &i_offset, &prevoff,
441 &enduseful, &ss);
442 if (error != 0) {
443 brelse(bp);
444 return (error);
445 }
446 if (entry_found) {
447 ep = (struct ext2fs_direct_2 *)((char *)bp->b_data +
448 (entryoffsetinblock & bmask));
449 foundentry:
450 ino = ep->e2d_ino;
451 goto found;
452 }
453 }
454 notfound:
455 /*
456 * If we started in the middle of the directory and failed
457 * to find our target, we must check the beginning as well.
458 */
459 if (numdirpasses == 2) {
460 numdirpasses--;
461 i_offset = 0;
462 endsearch = i_diroff;
463 goto searchloop;
464 }
465 if (bp != NULL)
466 brelse(bp);
467 /*
468 * If creating, and at end of pathname and current
469 * directory has not been removed, then can consider
470 * allowing file to be created.
471 */
472 if ((nameiop == CREATE || nameiop == RENAME) &&
473 (flags & ISLASTCN) && dp->i_nlink != 0) {
474 /*
475 * Access for write is interpreted as allowing
476 * creation of files in the directory.
477 */
478 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0)
479 return (error);
480 /*
481 * Return an indication of where the new directory
482 * entry should be put. If we didn't find a slot,
483 * then set dp->i_count to 0 indicating
484 * that the new slot belongs at the end of the
485 * directory. If we found a slot, then the new entry
486 * can be put in the range from dp->i_offset to
487 * dp->i_offset + dp->i_count.
488 */
489 if (ss.slotstatus == NONE) {
490 dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
491 dp->i_count = 0;
492 enduseful = dp->i_offset;
493 } else {
494 dp->i_offset = ss.slotoffset;
495 dp->i_count = ss.slotsize;
496 if (enduseful < ss.slotoffset + ss.slotsize)
497 enduseful = ss.slotoffset + ss.slotsize;
498 }
499 dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
500 /*
501 * We return with the directory locked, so that
502 * the parameters we set up above will still be
503 * valid if we actually decide to do a direnter().
504 * We return ni_vp == NULL to indicate that the entry
505 * does not currently exist; we leave a pointer to
506 * the (locked) directory inode in ndp->ni_dvp.
507 * The pathname buffer is saved so that the name
508 * can be obtained later.
509 *
510 * NB - if the directory is unlocked, then this
511 * information cannot be used.
512 */
513 cnp->cn_flags |= SAVENAME;
514 return (EJUSTRETURN);
515 }
516 /*
517 * Insert name into cache (as non-existent) if appropriate.
518 */
519 if ((cnp->cn_flags & MAKEENTRY) != 0)
520 cache_enter(vdp, NULL, cnp);
521 return (ENOENT);
522
523 found:
524 if (dd_ino != NULL)
525 *dd_ino = ino;
526 if (numdirpasses == 2)
527 nchstats.ncs_pass2++;
528 /*
529 * Check that directory length properly reflects presence
530 * of this entry.
531 */
532 if (entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen)
533 > dp->i_size) {
534 ext2_dirbad(dp, i_offset, "i_size too small");
535 dp->i_size = entryoffsetinblock + EXT2_DIR_REC_LEN(ep->e2d_namlen);
536 dp->i_flag |= IN_CHANGE | IN_UPDATE;
537 }
538 brelse(bp);
539
540 /*
541 * Found component in pathname.
542 * If the final component of path name, save information
543 * in the cache as to where the entry was found.
544 */
545 if ((flags & ISLASTCN) && nameiop == LOOKUP)
546 dp->i_diroff = i_offset &~ (DIRBLKSIZ - 1);
547 /*
548 * If deleting, and at end of pathname, return
549 * parameters which can be used to remove file.
550 */
551 if (nameiop == DELETE && (flags & ISLASTCN)) {
552 if (flags & LOCKPARENT)
553 ASSERT_VOP_ELOCKED(vdp, __FUNCTION__);
554 /*
555 * Write access to directory required to delete files.
556 */
557 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0)
558 return (error);
559 /*
560 * Return pointer to current entry in dp->i_offset,
561 * and distance past previous entry (if there
562 * is a previous entry in this block) in dp->i_count.
563 * Save directory inode pointer in ndp->ni_dvp for dirremove().
564 *
565 * Technically we shouldn't be setting these in the
566 * WANTPARENT case (first lookup in rename()), but any
567 * lookups that will result in directory changes will
568 * overwrite these.
569 */
570 dp->i_offset = i_offset;
571 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
572 dp->i_count = 0;
573 else
574 dp->i_count = dp->i_offset - prevoff;
575 if (dd_ino != NULL)
576 return (0);
577 if (dp->i_number == ino) {
578 VREF(vdp);
579 *vpp = vdp;
580 return (0);
581 }
582 if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE,
583 &tdp)) != 0)
584 return (error);
585 /*
586 * If directory is "sticky", then user must own
587 * the directory, or the file in it, else she
588 * may not delete it (unless she's root). This
589 * implements append-only directories.
590 */
591 if ((dp->i_mode & ISVTX) &&
592 cred->cr_uid != 0 &&
593 cred->cr_uid != dp->i_uid &&
594 VTOI(tdp)->i_uid != cred->cr_uid) {
595 vput(tdp);
596 return (EPERM);
597 }
598 *vpp = tdp;
599 return (0);
600 }
601
602 /*
603 * If rewriting (RENAME), return the inode and the
604 * information required to rewrite the present directory
605 * Must get inode of directory entry to verify it's a
606 * regular file, or empty directory.
607 */
608 if (nameiop == RENAME && (flags & ISLASTCN)) {
609 if ((error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread)) != 0)
610 return (error);
611 /*
612 * Careful about locking second inode.
613 * This can only occur if the target is ".".
614 */
615 dp->i_offset = i_offset;
616 if (dp->i_number == ino)
617 return (EISDIR);
618 if (dd_ino != NULL)
619 return (0);
620 if ((error = VFS_VGET(vdp->v_mount, ino, LK_EXCLUSIVE,
621 &tdp)) != 0)
622 return (error);
623 *vpp = tdp;
624 cnp->cn_flags |= SAVENAME;
625 return (0);
626 }
627 if (dd_ino != NULL)
628 return (0);
629
630 /*
631 * Step through the translation in the name. We do not `vput' the
632 * directory because we may need it again if a symbolic link
633 * is relative to the current directory. Instead we save it
634 * unlocked as "pdp". We must get the target inode before unlocking
635 * the directory to insure that the inode will not be removed
636 * before we get it. We prevent deadlock by always fetching
637 * inodes from the root, moving down the directory tree. Thus
638 * when following backward pointers ".." we must unlock the
639 * parent directory before getting the requested directory.
640 * There is a potential race condition here if both the current
641 * and parent directories are removed before the VFS_VGET for the
642 * inode associated with ".." returns. We hope that this occurs
643 * infrequently since we cannot avoid this race condition without
644 * implementing a sophisticated deadlock detection algorithm.
645 * Note also that this simple deadlock detection scheme will not
646 * work if the file system has any hard links other than ".."
647 * that point backwards in the directory structure.
648 */
649 pdp = vdp;
650 if (flags & ISDOTDOT) {
651 error = vn_vget_ino(pdp, ino, cnp->cn_lkflags, &tdp);
652 if (pdp->v_iflag & VI_DOOMED) {
653 if (error == 0)
654 vput(tdp);
655 error = ENOENT;
656 }
657 if (error)
658 return (error);
659 /*
660 * Recheck that ".." entry in the vdp directory points
661 * to the inode we looked up before vdp lock was
662 * dropped.
663 */
664 error = ext2_lookup_ino(pdp, NULL, cnp, &ino1);
665 if (error) {
666 vput(tdp);
667 return (error);
668 }
669 if (ino1 != ino) {
670 vput(tdp);
671 goto restart;
672 }
673 *vpp = tdp;
674 } else if (dp->i_number == ino) {
675 VREF(vdp); /* we want ourself, ie "." */
676 /*
677 * When we lookup "." we still can be asked to lock it
678 * differently.
679 */
680 ltype = cnp->cn_lkflags & LK_TYPE_MASK;
681 if (ltype != VOP_ISLOCKED(vdp)) {
682 if (ltype == LK_EXCLUSIVE)
683 vn_lock(vdp, LK_UPGRADE | LK_RETRY);
684 else /* if (ltype == LK_SHARED) */
685 vn_lock(vdp, LK_DOWNGRADE | LK_RETRY);
686 }
687 *vpp = vdp;
688 } else {
689 if ((error = VFS_VGET(vdp->v_mount, ino, cnp->cn_lkflags,
690 &tdp)) != 0)
691 return (error);
692 *vpp = tdp;
693 }
694
695 /*
696 * Insert name into cache if appropriate.
697 */
698 if (cnp->cn_flags & MAKEENTRY)
699 cache_enter(vdp, *vpp, cnp);
700 return (0);
701 }
702
703 int
704 ext2_search_dirblock(struct inode *ip, void *data, int *foundp,
705 const char *name, int namelen, int *entryoffsetinblockp,
706 doff_t *offp, doff_t *prevoffp, doff_t *endusefulp,
707 struct ext2fs_searchslot *ssp)
708 {
709 struct vnode *vdp;
710 struct ext2fs_direct_2 *ep, *top;
711 uint32_t bsize = ip->i_e2fs->e2fs_bsize;
712 int offset = *entryoffsetinblockp;
713 int namlen;
714
715 vdp = ITOV(ip);
716
717 ep = (struct ext2fs_direct_2 *)((char *)data + offset);
718 top = (struct ext2fs_direct_2 *)((char *)data +
719 bsize - EXT2_DIR_REC_LEN(0));
720
721 while (ep < top) {
722 /*
723 * Full validation checks are slow, so we only check
724 * enough to insure forward progress through the
725 * directory. Complete checks can be run by setting
726 * "vfs.e2fs.dirchk" to be true.
727 */
728 if (ep->e2d_reclen == 0 ||
729 (dirchk && ext2_dirbadentry(vdp, ep, offset))) {
730 int i;
731
732 ext2_dirbad(ip, *offp, "mangled entry");
733 i = bsize - (offset & (bsize - 1));
734 *offp += i;
735 offset += i;
736 continue;
737 }
738
739 /*
740 * If an appropriate sized slot has not yet been found,
741 * check to see if one is available. Also accumulate space
742 * in the current block so that we can determine if
743 * compaction is viable.
744 */
745 if (ssp->slotstatus != FOUND) {
746 int size = ep->e2d_reclen;
747
748 if (ep->e2d_ino != 0)
749 size -= EXT2_DIR_REC_LEN(ep->e2d_namlen);
750 if (size > 0) {
751 if (size >= ssp->slotneeded) {
752 ssp->slotstatus = FOUND;
753 ssp->slotoffset = *offp;
754 ssp->slotsize = ep->e2d_reclen;
755 } else if (ssp->slotstatus == NONE) {
756 ssp->slotfreespace += size;
757 if (ssp->slotoffset == -1)
758 ssp->slotoffset = *offp;
759 if (ssp->slotfreespace >= ssp->slotneeded) {
760 ssp->slotstatus = COMPACT;
761 ssp->slotsize = *offp +
762 ep->e2d_reclen -
763 ssp->slotoffset;
764 }
765 }
766 }
767 }
768 /*
769 * Check for a name match.
770 */
771 if (ep->e2d_ino) {
772 namlen = ep->e2d_namlen;
773 if (namlen == namelen &&
774 !bcmp(name, ep->e2d_name, (unsigned)namlen)) {
775 /*
776 * Save directory entry's inode number and
777 * reclen in ndp->ni_ufs area, and release
778 * directory buffer.
779 */
780 *foundp = 1;
781 return (0);
782 }
783 }
784 *prevoffp = *offp;
785 *offp += ep->e2d_reclen;
786 offset += ep->e2d_reclen;
787 *entryoffsetinblockp = offset;
788 if (ep->e2d_ino)
789 *endusefulp = *offp;
790 /*
791 * Get pointer to the next entry.
792 */
793 ep = (struct ext2fs_direct_2 *)((char *)data + offset);
794 }
795
796 return (0);
797 }
798
799 void
800 ext2_dirbad(struct inode *ip, doff_t offset, char *how)
801 {
802 struct mount *mp;
803
804 mp = ITOV(ip)->v_mount;
805 if ((mp->mnt_flag & MNT_RDONLY) == 0)
806 panic("ext2_dirbad: %s: bad dir ino %lu at offset %ld: %s\n",
807 mp->mnt_stat.f_mntonname, (u_long)ip->i_number,(long)offset, how);
808 else
809 (void)printf("%s: bad dir ino %lu at offset %ld: %s\n",
810 mp->mnt_stat.f_mntonname, (u_long)ip->i_number, (long)offset, how);
811
812 }
813
814 /*
815 * Do consistency checking on a directory entry:
816 * record length must be multiple of 4
817 * entry must fit in rest of its DIRBLKSIZ block
818 * record must be large enough to contain entry
819 * name is not longer than MAXNAMLEN
820 * name must be as long as advertised, and null terminated
821 */
822 /*
823 * changed so that it confirms to ext2_check_dir_entry
824 */
825 static int
826 ext2_dirbadentry(struct vnode *dp, struct ext2fs_direct_2 *de,
827 int entryoffsetinblock)
828 {
829 int DIRBLKSIZ = VTOI(dp)->i_e2fs->e2fs_bsize;
830
831 char *error_msg = NULL;
832
833 if (de->e2d_reclen < EXT2_DIR_REC_LEN(1))
834 error_msg = "rec_len is smaller than minimal";
835 else if (de->e2d_reclen % 4 != 0)
836 error_msg = "rec_len % 4 != 0";
837 else if (de->e2d_reclen < EXT2_DIR_REC_LEN(de->e2d_namlen))
838 error_msg = "reclen is too small for name_len";
839 else if (entryoffsetinblock + de->e2d_reclen > DIRBLKSIZ)
840 error_msg = "directory entry across blocks";
841 /* else LATER
842 if (de->inode > dir->i_sb->u.ext2_sb.s_es->s_inodes_count)
843 error_msg = "inode out of bounds";
844 */
845
846 if (error_msg != NULL) {
847 printf("bad directory entry: %s\n", error_msg);
848 printf("offset=%d, inode=%lu, rec_len=%u, name_len=%u\n",
849 entryoffsetinblock, (unsigned long)de->e2d_ino,
850 de->e2d_reclen, de->e2d_namlen);
851 }
852 return error_msg == NULL ? 0 : 1;
853 }
854
855 /*
856 * Write a directory entry after a call to namei, using the parameters
857 * that it left in nameidata. The argument ip is the inode which the new
858 * directory entry will refer to. Dvp is a pointer to the directory to
859 * be written, which was left locked by namei. Remaining parameters
860 * (dp->i_offset, dp->i_count) indicate how the space for the new
861 * entry is to be obtained.
862 */
863 int
864 ext2_direnter(struct inode *ip, struct vnode *dvp, struct componentname *cnp)
865 {
866 struct inode *dp;
867 struct ext2fs_direct_2 newdir;
868 struct iovec aiov;
869 struct uio auio;
870 int error, newentrysize;
871 int DIRBLKSIZ = ip->i_e2fs->e2fs_bsize;
872
873
874 #ifdef INVARIANTS
875 if ((cnp->cn_flags & SAVENAME) == 0)
876 panic("ext2_direnter: missing name");
877 #endif
878 dp = VTOI(dvp);
879 newdir.e2d_ino = ip->i_number;
880 newdir.e2d_namlen = cnp->cn_namelen;
881 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
882 EXT2F_INCOMPAT_FTYPE))
883 newdir.e2d_type = DTTOFT(IFTODT(ip->i_mode));
884 else
885 newdir.e2d_type = EXT2_FT_UNKNOWN;
886 bcopy(cnp->cn_nameptr, newdir.e2d_name, (unsigned)cnp->cn_namelen + 1);
887 newentrysize = EXT2_DIR_REC_LEN(newdir.e2d_namlen);
888
889 if (ext2_htree_has_idx(dp)) {
890 error = ext2_htree_add_entry(dvp, &newdir, cnp);
891 if (error) {
892 dp->i_flag &= ~IN_E3INDEX;
893 dp->i_flag |= IN_CHANGE | IN_UPDATE;
894 }
895 return (error);
896 }
897
898 if (EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_DIRHASHINDEX) &&
899 !ext2_htree_has_idx(dp)) {
900 if ((dp->i_size / DIRBLKSIZ) == 1 &&
901 dp->i_offset == DIRBLKSIZ) {
902 /*
903 * Making indexed directory when one block is not
904 * enough to save all entries.
905 */
906 return ext2_htree_create_index(dvp, cnp, &newdir);
907 }
908 }
909
910 if (dp->i_count == 0) {
911 /*
912 * If dp->i_count is 0, then namei could find no
913 * space in the directory. Here, dp->i_offset will
914 * be on a directory block boundary and we will write the
915 * new entry into a fresh block.
916 */
917 if (dp->i_offset & (DIRBLKSIZ - 1))
918 panic("ext2_direnter: newblk");
919 auio.uio_offset = dp->i_offset;
920 newdir.e2d_reclen = DIRBLKSIZ;
921 auio.uio_resid = newentrysize;
922 aiov.iov_len = newentrysize;
923 aiov.iov_base = (caddr_t)&newdir;
924 auio.uio_iov = &aiov;
925 auio.uio_iovcnt = 1;
926 auio.uio_rw = UIO_WRITE;
927 auio.uio_segflg = UIO_SYSSPACE;
928 auio.uio_td = (struct thread *)0;
929 error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
930 if (DIRBLKSIZ >
931 VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
932 /* XXX should grow with balloc() */
933 panic("ext2_direnter: frag size");
934 else if (!error) {
935 dp->i_size = roundup2(dp->i_size, DIRBLKSIZ);
936 dp->i_flag |= IN_CHANGE;
937 }
938 return (error);
939 }
940
941 error = ext2_add_entry(dvp, &newdir);
942 if (!error && dp->i_endoff && dp->i_endoff < dp->i_size)
943 error = ext2_truncate(dvp, (off_t)dp->i_endoff, IO_SYNC,
944 cnp->cn_cred, cnp->cn_thread);
945 return (error);
946 }
947
948 /*
949 * Insert an entry into the directory block.
950 * Compact the contents.
951 */
952 int
953 ext2_add_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry)
954 {
955 struct ext2fs_direct_2 *ep, *nep;
956 struct inode *dp;
957 struct buf *bp;
958 u_int dsize;
959 int error, loc, newentrysize, spacefree;
960 char *dirbuf;
961
962 dp = VTOI(dvp);
963
964 /*
965 * If dp->i_count is non-zero, then namei found space
966 * for the new entry in the range dp->i_offset to
967 * dp->i_offset + dp->i_count in the directory.
968 * To use this space, we may have to compact the entries located
969 * there, by copying them together towards the beginning of the
970 * block, leaving the free space in one usable chunk at the end.
971 */
972
973 /*
974 * Increase size of directory if entry eats into new space.
975 * This should never push the size past a new multiple of
976 * DIRBLKSIZE.
977 *
978 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
979 */
980 if (dp->i_offset + dp->i_count > dp->i_size)
981 dp->i_size = dp->i_offset + dp->i_count;
982 /*
983 * Get the block containing the space for the new directory entry.
984 */
985 if ((error = ext2_blkatoff(dvp, (off_t)dp->i_offset, &dirbuf,
986 &bp)) != 0)
987 return (error);
988 /*
989 * Find space for the new entry. In the simple case, the entry at
990 * offset base will have the space. If it does not, then namei
991 * arranged that compacting the region dp->i_offset to
992 * dp->i_offset + dp->i_count would yield the
993 * space.
994 */
995 newentrysize = EXT2_DIR_REC_LEN(entry->e2d_namlen);
996 ep = (struct ext2fs_direct_2 *)dirbuf;
997 dsize = EXT2_DIR_REC_LEN(ep->e2d_namlen);
998 spacefree = ep->e2d_reclen - dsize;
999 for (loc = ep->e2d_reclen; loc < dp->i_count; ) {
1000 nep = (struct ext2fs_direct_2 *)(dirbuf + loc);
1001 if (ep->e2d_ino) {
1002 /* trim the existing slot */
1003 ep->e2d_reclen = dsize;
1004 ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
1005 } else {
1006 /* overwrite; nothing there; header is ours */
1007 spacefree += dsize;
1008 }
1009 dsize = EXT2_DIR_REC_LEN(nep->e2d_namlen);
1010 spacefree += nep->e2d_reclen - dsize;
1011 loc += nep->e2d_reclen;
1012 bcopy((caddr_t)nep, (caddr_t)ep, dsize);
1013 }
1014 /*
1015 * Update the pointer fields in the previous entry (if any),
1016 * copy in the new entry, and write out the block.
1017 */
1018 if (ep->e2d_ino == 0) {
1019 if (spacefree + dsize < newentrysize)
1020 panic("ext2_direnter: compact1");
1021 entry->e2d_reclen = spacefree + dsize;
1022 } else {
1023 if (spacefree < newentrysize)
1024 panic("ext2_direnter: compact2");
1025 entry->e2d_reclen = spacefree;
1026 ep->e2d_reclen = dsize;
1027 ep = (struct ext2fs_direct_2 *)((char *)ep + dsize);
1028 }
1029 bcopy((caddr_t)entry, (caddr_t)ep, (u_int)newentrysize);
1030 if (DOINGASYNC(dvp)) {
1031 bdwrite(bp);
1032 error = 0;
1033 } else {
1034 error = bwrite(bp);
1035 }
1036 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1037 return (error);
1038 }
1039
1040 /*
1041 * Remove a directory entry after a call to namei, using
1042 * the parameters which it left in nameidata. The entry
1043 * dp->i_offset contains the offset into the directory of the
1044 * entry to be eliminated. The dp->i_count field contains the
1045 * size of the previous record in the directory. If this
1046 * is 0, the first entry is being deleted, so we need only
1047 * zero the inode number to mark the entry as free. If the
1048 * entry is not the first in the directory, we must reclaim
1049 * the space of the now empty record by adding the record size
1050 * to the size of the previous entry.
1051 */
1052 int
1053 ext2_dirremove(struct vnode *dvp, struct componentname *cnp)
1054 {
1055 struct inode *dp;
1056 struct ext2fs_direct_2 *ep, *rep;
1057 struct buf *bp;
1058 int error;
1059
1060 dp = VTOI(dvp);
1061 if (dp->i_count == 0) {
1062 /*
1063 * First entry in block: set d_ino to zero.
1064 */
1065 if ((error =
1066 ext2_blkatoff(dvp, (off_t)dp->i_offset, (char **)&ep,
1067 &bp)) != 0)
1068 return (error);
1069 ep->e2d_ino = 0;
1070 error = bwrite(bp);
1071 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1072 return (error);
1073 }
1074 /*
1075 * Collapse new free space into previous entry.
1076 */
1077 if ((error = ext2_blkatoff(dvp, (off_t)(dp->i_offset - dp->i_count),
1078 (char **)&ep, &bp)) != 0)
1079 return (error);
1080
1081 /* Set 'rep' to the entry being removed. */
1082 if (dp->i_count == 0)
1083 rep = ep;
1084 else
1085 rep = (struct ext2fs_direct_2 *)((char *)ep + ep->e2d_reclen);
1086 ep->e2d_reclen += rep->e2d_reclen;
1087 if (DOINGASYNC(dvp) && dp->i_count != 0)
1088 bdwrite(bp);
1089 else
1090 error = bwrite(bp);
1091 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1092 return (error);
1093 }
1094
1095 /*
1096 * Rewrite an existing directory entry to point at the inode
1097 * supplied. The parameters describing the directory entry are
1098 * set up by a call to namei.
1099 */
1100 int
1101 ext2_dirrewrite(struct inode *dp, struct inode *ip, struct componentname *cnp)
1102 {
1103 struct buf *bp;
1104 struct ext2fs_direct_2 *ep;
1105 struct vnode *vdp = ITOV(dp);
1106 int error;
1107
1108 if ((error = ext2_blkatoff(vdp, (off_t)dp->i_offset, (char **)&ep,
1109 &bp)) != 0)
1110 return (error);
1111 ep->e2d_ino = ip->i_number;
1112 if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs,
1113 EXT2F_INCOMPAT_FTYPE))
1114 ep->e2d_type = DTTOFT(IFTODT(ip->i_mode));
1115 else
1116 ep->e2d_type = EXT2_FT_UNKNOWN;
1117 error = bwrite(bp);
1118 dp->i_flag |= IN_CHANGE | IN_UPDATE;
1119 return (error);
1120 }
1121
1122 /*
1123 * Check if a directory is empty or not.
1124 * Inode supplied must be locked.
1125 *
1126 * Using a struct dirtemplate here is not precisely
1127 * what we want, but better than using a struct direct.
1128 *
1129 * NB: does not handle corrupted directories.
1130 */
1131 int
1132 ext2_dirempty(struct inode *ip, ino_t parentino, struct ucred *cred)
1133 {
1134 off_t off;
1135 struct dirtemplate dbuf;
1136 struct ext2fs_direct_2 *dp = (struct ext2fs_direct_2 *)&dbuf;
1137 int error, namlen;
1138 ssize_t count;
1139 #define MINDIRSIZ (sizeof(struct dirtemplate) / 2)
1140
1141 for (off = 0; off < ip->i_size; off += dp->e2d_reclen) {
1142 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ,
1143 off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred,
1144 NOCRED, &count, (struct thread *)0);
1145 /*
1146 * Since we read MINDIRSIZ, residual must
1147 * be 0 unless we're at end of file.
1148 */
1149 if (error || count != 0)
1150 return (0);
1151 /* avoid infinite loops */
1152 if (dp->e2d_reclen == 0)
1153 return (0);
1154 /* skip empty entries */
1155 if (dp->e2d_ino == 0)
1156 continue;
1157 /* accept only "." and ".." */
1158 namlen = dp->e2d_namlen;
1159 if (namlen > 2)
1160 return (0);
1161 if (dp->e2d_name[0] != '.')
1162 return (0);
1163 /*
1164 * At this point namlen must be 1 or 2.
1165 * 1 implies ".", 2 implies ".." if second
1166 * char is also "."
1167 */
1168 if (namlen == 1)
1169 continue;
1170 if (dp->e2d_name[1] == '.' && dp->e2d_ino == parentino)
1171 continue;
1172 return (0);
1173 }
1174 return (1);
1175 }
1176
1177 /*
1178 * Check if source directory is in the path of the target directory.
1179 * Target is supplied locked, source is unlocked.
1180 * The target is always vput before returning.
1181 */
1182 int
1183 ext2_checkpath(struct inode *source, struct inode *target, struct ucred *cred)
1184 {
1185 struct vnode *vp;
1186 int error, namlen;
1187 struct dirtemplate dirbuf;
1188
1189 vp = ITOV(target);
1190 if (target->i_number == source->i_number) {
1191 error = EEXIST;
1192 goto out;
1193 }
1194 if (target->i_number == EXT2_ROOTINO) {
1195 error = 0;
1196 goto out;
1197 }
1198
1199 for (;;) {
1200 if (vp->v_type != VDIR) {
1201 error = ENOTDIR;
1202 break;
1203 }
1204 error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
1205 sizeof(struct dirtemplate), (off_t)0, UIO_SYSSPACE,
1206 IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, NULL,
1207 NULL);
1208 if (error != 0)
1209 break;
1210 namlen = dirbuf.dotdot_type; /* like ufs little-endian */
1211 if (namlen != 2 ||
1212 dirbuf.dotdot_name[0] != '.' ||
1213 dirbuf.dotdot_name[1] != '.') {
1214 error = ENOTDIR;
1215 break;
1216 }
1217 if (dirbuf.dotdot_ino == source->i_number) {
1218 error = EINVAL;
1219 break;
1220 }
1221 if (dirbuf.dotdot_ino == EXT2_ROOTINO)
1222 break;
1223 vput(vp);
1224 if ((error = VFS_VGET(vp->v_mount, dirbuf.dotdot_ino,
1225 LK_EXCLUSIVE, &vp)) != 0) {
1226 vp = NULL;
1227 break;
1228 }
1229 }
1230
1231 out:
1232 if (error == ENOTDIR)
1233 printf("checkpath: .. not a directory\n");
1234 if (vp != NULL)
1235 vput(vp);
1236 return (error);
1237 }
Cache object: 4c4c2aeeaf016d21b5750ba5bd46c487
|